Transcript
Hi Mike Matchett with Small World Big Data and we are here today talking about AI of course, but really a practical and validating use of AI. We're talking with Tray.ai, and we'll be going into some things they're doing to help you create agents and specific work domains and, and make those agents interrelate and really increase your automation game. Uh, in the topic of iPads, if you're there. Uh, we're going to see how you start to fold in AI and agents to be totally more productive. So just hold on a second. Hey, Rich. Uh, welcome to our show today. Thanks for having me. All right, so you are, uh, deeply involved with Tray.ai, and this is, uh, something that you got involved with, uh, and kind of in this iPad space. Maybe you could explain, uh, a little bit about how you came to be involved in this integration automation end of the IT market. Yeah, I guess like any good founder, the company was started out of frustration. Um, the the sort of vision that we had initially was what if you could take your sort of smart, technically capable, uh, employee within an organization and give them the capability of a programmer? Because all of the applications we work with, all of the services that we get involved with, you know, the true power is unlocked through APIs and access. And so our approach to integration and ipaas was really built around that. Can we make this easy? Can we make this powerful? Can we make it so that it can consume the huge workloads that come with cloud and AI power? I mean, AI just as a separate sort of line of thought as an analyst, AI seems greatly aligned with increasing automation and doing integration activities and stuff. So I think that's that's really a great use case. Uh, just maybe for folks, we talk about this term ipaas. What is that just real quick. Yeah. Integration platform as a service. Okay, okay. So that's, uh, more cloud based then. Right. You're you're you're you're taking things that might have been done on premise and coming at it from a, I guess, a more of a global integration perspective. Yeah. I guess Ipaas is a kind of generalist term for integration platforms. And if you sort of go back to the start of software, um, the challenges that I passes solve have changed over that course. So, you know, large ERP on prem to on prem, on prem to cloud, cloud to cloud. And now with AI it's a new integration challenge. And that's really the evolution of Ipaas today. All right. So we put AI at the end of that thing. Certainly everyone's doing it. Um let's talk a little bit about uh, you know what what, uh, Tray does before we get to AI, uh, we talk about integrations. So you just mentioned, like, CRMs and HR and things. What what what should someone think of as an integration activity? Yeah. So, Tray, um, is known for being a enterprise grade integration, integration and automation provider. So we provide a platform that allows you to connect into any service, any database, anything that has an endpoint essentially, and create integration or automation around it. So traditionally that might be, you know, there are 30 different systems where the, the, the trigger point from one means that you want to pull data, manipulate it in some way, do something with it downstream in certain systems, and then output it to a service like Slack or Teams or something in between. So integrations can, in some cases seem like mini applications or full blown processes, but effectively they're able to carry out automation between lots of different functions and services. To make everything work better together. Basically, you invested here and invested there, combining combining the automations across it. So it does seem like for for a long time it's been very deterministic. Like I'm going to write a script or do some drag and drop thing and it's, it's it's got a specific workflow, a specific. But now we introduce AI to this and and people say like, hey, you know, I can ask ChatGPT or whatever AI I happen to have, uh, Gemini or Cloud, um, questions in natural language and have it have it, uh, answer me in that. And it seems like this is an area ripe for that kind of integration activity. Like, I don't I don't want something so deterministic. It's already just doing what it's doing. And it takes me an hour and a half to figure out what that is. But I just want to ask it something and have it go do it. So how did you guys see this coming on? And, and, uh, what do you think is sort of this bigger response to that? Yeah, I think really, you know, the, the if you, if you go back to the start for us, it was how do you get the machines to do the work. Right. Like a lot of the things that people have to do day to day are manual and repetitive. And the idea of automation is that you take those things away. So we were constantly looking at, you know, how can we use machine learning? How can we predict the next step? How can we even figure out what integrations should exist based on system access? And so is as AI kind of evolved, it really was was evident that all of the processes within a company are going to change. They're going to become, in some cases, you know, fully agentic and and in some cases they'll be, you know, um, more of a deterministic approach which uses AI where there might be part of a process where the agentic experience is important, but actually you still want it to be relatively deterministic. And I think for us, very early on, we looked at how can we evolve the power that we have to orchestrate and connect and join that up with the reasoning engine that AI provides. And so how that kind of comes together is ultimately the limiting factor to an agent is what knowledge it has access to, and what downstream services or what actions it can go and take. And so really, an orchestration engine like tray, um, provides a lot of richness on either side to give you a far more capable, uh, agentic workflow or agent experience. Right. And when you talk now about, uh, this platform. These are all the things that would have been built up for compliance and governance and, uh, making things work at scale and speed. Right? It's not just about making integration work, but making it bulletproof for a global enterprise to really trust this thing. Right? So, yeah. I mean, the hard part about integrations or even building initially with AI isn't necessarily writing the code to do the connectivity. It's everything that comes over the fold. It's how do you handle these things at scale? How do you handle, um, an API failing or a throughput challenge? How do you handle change over time? How do you handle logs and observability and governance? All of these things that have to be built out to build a mature solution. That's really where I think the power of these platforms comes to the fore. All right. And so so how do we fold AI into this? I mean, when I, when I think of when I think of automation, I'm thinking like, oh, I identified this transaction has to happen. And then I got a lot of if then else kinds of statements or if this then that kinds of structures. But now I can build an agent. But that seems like I've just stepped off a cliff into into into thin air. What is what is building an agent look like? Yeah, I think I mean, you know, I'm going to give you two examples. There are many more. But I think in a lot of cases, um, people try and apply an agent to a solution that maybe doesn't need one. Right. So I think what we've seen is a lot of people build out what we call a generic workflow, whereby maybe you have a, uh, quote to cash integration within your finance and go to market operations, and there might be a component of that where you want to, um, use AI to pull data from an invoice, semantically analyze it and do something with it. And you can kind of isolate that, that, that subsection of the workflow, which may have required human interaction and keep that in the in the for with a more deterministic outcome. So we see a lot of that kind of infusing AI into existing workflows. And then secondarily you see a straight shot agent. So the example there would be uh like an ITSM agent. So you've got an IT support sort of question coming in. You know, that's essentially a prompt. What you're able to then do a tray is you can build an agent around that. And because tray can access any downstream system, can take action in any system. Maybe it's going to check something out in Okta, maybe it's going to use to access a user's machine. Maybe it's going to access knowledge from an internal knowledge base. It's it's able to provide a great deal of power in not just responding back to the user, but actually going and taking action. All right. So, uh, we talk about service management as being one of those really nice domains where we understand what's going on. Um, what are what are some other domains that we might think of within an organization where we can really get our hands around it? Yeah, we see lots of different examples. Um, you know, we see content generation engines in marketing. We see like automated, uh, like cool preps or cool research for salespeople. Um, we see HR agents that are able to, um, you know, process, um, change requests or handle policy questions or do the onboarding and offboarding. I think, you know, in its first instance, we're seeing a lot of people look at existing processes that they have where there's a back and forth interaction and looking at sort of making that agentic in some form. And so there's there's quite a lot of breadth in the way in which these solutions get applied. But there are probably some core starting points which are typically around the support use cases or the the areas where people are already used to having some, some kind of automation in place. And since since Tray, you know, already has what 700 some integrations under your belt bringing AI Agentry to this. I'm going to make up a new term here. Making making it Agentic. Uh, seems like it's a pretty natural evolution, because you can use that entire library of sort of constrained examples of to, to bring new agentic behavior to the board. Uh, but I have a question about sort of I don't know if it's if it's complexity or governance. Um, how do you how do you if someone starts building agents, they start accessing things everywhere. How do I how do I stay in control of that as an IT person or as a business person? Yeah. I mean, so I love the term agent, an agent. And I think we'll steal that as the, the sort of wizards of it. But, um, how people stay in control is a is a really important topic. And I think there are a few ways to do it. So firstly, it's building agents within domains because they're more effective when they're focused. So having this idea of multiple agents working together allows you to to sort of put additional governance around what an agent does. Secondarily, you can put controls and guardrails around what your agents do. So you can look for certain inputs and you can react to those and, you know, tokenize sensitive data and do something with it. All of that you're able to do kind of within the tray platform. And so often we're seeing teams, you know, start out with, um, what they want the agent to do, rolling it out within test groups, um, kind of constantly adjusting the governance and controls that they put around it. And, um, and ultimately that's the thing that that matters probably most for a, for a company to to get on this journey. And and while they're doing that, I mean, just, just technically, uh, you know, when agents want to get access to things. Uh, yeah, there's, you know, model control protocol. There's there's all sorts of other stuff, but but Tray is diving deep here with some new capabilities around those control points. Can you explain those a little bit? Yeah. So we've launched a product called Agent Gateway. And this provides like a control plane for our customers to be able to, uh, provide controlled access to MCP servers, um, manage authentications to those various servers, uh, expose tools created entre um via MCP to third parties. And you can kind of think of it as the barrier that sits between these different services and ensures that the person who's trying to access it has the authentication, authentication credentials that they need, that the, uh, the access to the tools required is, is limited to the scope that is necessary for, for that individual. Without that, it's very hard to do these things. You know, using, uh, protocols like MCP often means that you're getting every tool that's available to that MCP server that you've connected to, and in some cases that can that can introduce risk. And so we provide our agent gateway capability because it allows you to to kind of manage that and manage the permissioning model of agents, which is is critically important. Yeah, it does scare me a little bit when I see what some people are doing with MCP. You know, I go in, I go and ask another copilot or something like write me an MCP server for this service, and then they post it. And internally in the organization. Yeah, they've just exposed everything, you know, to anybody who wants to validate server. How do you validate who owns MCP server. Right. I think we're still in the early innings of these things. So there's there really has to be a lot of caution taken when when it comes to connectivity integration. Uh, I'm sure there's some governance and compliance aspects to this, which a lot of people are going to be very concerned with. Um, you know, so it's not simply making it work at performance and scale, but also making sure it's well, well audited and provable in that sense. What would you say to someone who says, well, AI hallucinate and they make stuff up? But I have to. But now I'm giving all the keys to my kingdom, to these agents. And how do I how do I how do I really ensure that, you know, things aren't going to go haywire? Yeah, I think the devil's in the detail, and that's in how you construct the agent in the first place. So we encourage our customers to build in, um, like static checks within how the agent performs so that, for example, if a user says or a user tries to say, you know, I'm this person, it shouldn't rely on their response. It should be checking where did the response come from? Had they authenticated into that service? Like, can I use sort of multiple areas to actually circle around and figure out the identity of that individual and how and how they've been able to make that request? Same goes for accessing knowledge and permissioning. And I think it it is important kind of how you construct and what services you use to be able to to enable agents in the first place. Because otherwise, um, you know, hallucinations are a risk. And I think everybody's been slightly mis sold on this concept that, um, we can just throw all this unstructured data into a vector, and now it's, like magically available to everybody in a, in a, in a, you know, brilliantly, um, carefully permissioned model when, when really it isn't because people haven't historically sort of maintained their own data that way. So I think it is about how you siphon these things off and how you actually kind of provide access in the first place. That does make a difference. Yeah. It's like it's like you can create agents, but there'd be like children if everybody's just creating that, you have to have some some maturity process and actually create them and train them up a little bit so they become fully functioning. Yeah. Agent. Yeah. That's the analogy we use is when you start off with an agent, it's kind of like hiring an intern that has all the knowledge of the internet. You know, depending on what knowledge you've given it or what knowledge is available. But it's they aren't stuck for for knowing what to do. But they you got to train them up over time. You got to be able to kind of mold them toward, um, gaining the experience to, to do these things effectively. So it sounds like Tray is well ahead of this game, and especially in this integration, uh, domain, uh, that you've got lots of guardrails, you've got a gateway, you've got management, you've got ways to, uh, to to govern MCP interactions, turn all your agents, I believe, into MCP servers in turn, sort of automatically, uh, or when necessary. Right. So you people can build up layers of value, um, which is very cool. So what what's what's sort of next on your horizon? Where, where do you, where do you look down the road and say, like, what's going to happen in 2026 here. Yeah. So we're kind of aligned by the challenges we see our customers having, which primarily are IT audiences. And what we're hearing and seeing from them is, um, uh, AI orchestration and governance is the number one headache. So a lot of what we're working on is about providing, uh, more control, um, unlocking more power for these departments. Um, thinking about how we interact with the party agents and third party solutions and ultimately, um, all of the things that these departments historically looked for in their integration vendor when thinking about their integration roadmap, there's quite a lot of parity with how you think about deploying AI across across a company as well. So I think, you know, the holy grail for all of us is, uh, multiple agents that can be self-sufficient, that can carry out tasks on our behalf and, and enable us to do our very best work. And I think ultimately, getting putting our customers on that journey starts with giving them control, governance and the right access along the way. That's great power, right? That's it. Uh, if someone wants to learn more about this Rich, if someone wants to dig into trade, I want to look a little bit more at IPaths. They want to certainly hear about your AI Agentry stories. Uh, yeah. He's not going to work that term in. You got it on there. Well, you've got a website. Uh, anything specific you point them out there? Yeah. I think, um, it sounds like a bit of a no brainer, but our resources section is great. There's a bunch of customer stories on there. We regularly post interviews, um, on LinkedIn, which are our customers talking about exactly what they've built. And I find that that's most useful for people because a great part of the challenge is the inspiration and understanding the art of the possible. And so hearing what people have brought to life, um, from themselves, I think makes a huge difference. So, yeah, um, hit up our website, hit the resources section and follow us on LinkedIn for for more. All right. That is awesome. Rich, thank you for being here today. Thanks for explaining the AI story, at least introducing it to us. I'm sure there's a lot more to dive into. Uh, and, uh, check out, uh, check out AI if you all got integration challenges, we know it and governance challenges. So thanks again, Rich. Thanks, Mike. All right. Take care folks.
